Companies Are Just a Graph of Algorithms | Hacker News Reader

Companies Are Just a Graph of Algorithms

danielmiessler.com

29 pointssamuel24619 comments

When I see a complex socioeconomic phenomenon described as "just" I know for certain that someone is about to spectacularly fail to have read books.

This is value stream mapping. No, business process reengineering. No, systems dynamics. No, a Krebs cycle. No, ...

People could always do these things. It was never a sword that only AI enthusiasts could draw from the stone. By god people, the AI has read books, can't you give it a bash too?

Right. Because something is reducible, modulo a bunch of stuff you're not looking at, to a simplistic model, doesn't make it "just" that. Use of the word "just" is often a good tell you're about to read something naive.

How many times do we have to go through this? Humans invent a new technology and think it applies to everything!

Imagine the brain as a complex series of clockwork mechanisms…

Society can be modeled as a complex series of hydraulic tubes…

Companies are really a set of APIs between different departments…

Sure, these are all somewhat useful metaphors in context. But no one has built a working brain out of Lego. Sloshing water around to model an economy didn't produce unending wealth. Most companies aren't shuffling data around SOAP endpoints and winning capitalism.

Everyone seems to think AI is useful for someone else's problem, but not their own. Is a company a series of algorithms? I guess if you squint. Really it is a set of social dynamics,interpersonal relationships, and imperfect decisions.

Given that the AI companies themselves haven't replaced all their marketing departments, accountancy, and CEOs with AI - I guess the rest of us should probably wait.

Agreed that there's some measure of chaotic/creative work that won't fall into this type of category.

But much or most regular enterprise work is very much able to be done by having and regularly updating an SOP and then executing the task according to that SOP.

We suck at it. AI will be far better at it. And we'll sit above it and decide how to tweak the SOP based on taste/preference/expertise, whatever.

But the day-to-day work of handling insurance claims, doing procurement, doing analysis, creating reports, processing inputs according to some standard and producing some output according to another standard...that will largely be done by AI.

This is what makes the new /workflows feature coming to Claude Code so exciting (and frightening) to businesses. Along with Skills and Cowork and such, plus their analogs from other providers, Workflows are literally the making of opaque, alchemy-like work that Chris and Raj and Sarah do...into transparent, optimizable algorithms.

It's super hard to automate this stuff because it's super hard to articulate it. That's kind of the meta-super-power in all of this: the fact that AI is making the opaque and complex into transparent and inspectable.

I recognize all these words, I can sense, perceive, parse and reason about all of these words. Yet I cannot derive a sensible meaning from them. The pragmatics is of a flailing fish singing Waltzing Matilda to a teleporting cucumber.

There's nothing transparent or inspectable about a sparse fog of floating point numbers. Rendered as a picture it's the color of a television, tuned to a dead channel. As a sound: ksssshhhhhhhhh. If you see anything in it that you believe is a real discrete phenomenon then I have a face on Mars to sell you.

You know what is an inspectable algorithm? An algorithm. The old-fashioned kind that were intensional and not gigantic quasi-extensional stews connected to nervous cats in boxes. I'm so tired of this madness. So entirely bloody exhausted. Out of all the manias I've lived through in this trade the current one is by far the most absurd, wasteful and destructive.

Meanwhile thousands of people are using things like prompts and skills and Cowork to do actual work.

If you think this is theoretical at this point it's because you're not using it to do real stuff.

Mate until last week I had quite literally unlimited, unmetered access to frontier models from every major lab. No quotas, no brownouts, none of the stuff civilians gripe about. Chairman Mark had handed out the titanium amex and said "go forth and multiply my expenses".

What I saw was not ~ * ~ actual work ~ * ~. It wasn't "real stuff" either. What I witnessed was the most spectacular immolation of surplus I have ever seen. So much time, money and intelligence being pissed up against a wall in order to show the boss what a jolly good job you're doing.

When I tried to use it for actual work -- with, I repeat, utterly limitless amounts of tokens to burn -- it sucked balls. No matter how breathless the hype that this generation had finally cracked it, they all sucked. It produced flabby, buggy code by the gallon. It routinely fucked up and wasted my time. More than once I would wrestle with something for two or three days, give up, then bang a good solution in about an hour. Yes, I used rules and skills and .md files and and and and and and. Skill issue, you say? Well look, if hammers shot spikes through my hand one time in 20 then the skill issue is using the bloody doomhammer in the first place.

The actual work being done here is manic bullshitting and pissing in the village well. Shipping clanker clinker isn't productivity.

To reuse part of a 15-month old comment on LLM excitement:

> How might future generations visualize [our grand declarations]? I'm imagining some ancient Greeks, who have invented an inefficient reciprocating pump, which they declare is a heart and that means they've basically built a person. (At the time, many believed the brain was just there to cool the blood.) Look! The fluid being pumped can move a lever: It's waving to us.

Context: "The LLMentalist Effect " https://news.ycombinator.com/item?id=42983571

In defense of the OP, I have been reading something along the lines of "Limitation of liability, and the profit motive, have created AI in the form of Corporations, running distributed on human meatware, many years ago." for a decade or so. Its not an entirely baseless line of reasoning.

> AI excels at both discrete task execution and determining how things fit together, and every single one of your company's workflow components becomes ripe for optimization or elimination.

Why? On what basis is this claim made? They're trained to probabilistically complete patterns. Where does the confidence in this ability come from?

I personally like this article, its concepts, its charts and its graphs...

Observation: There probably is a market for documenting all of a business's processes and workflows (which business owner wouldn't want some really cool charts of all of their business processes?), and that should be able to done quickly and cheaply with Text-to-Image LLM's (NanoBanana, ?, ???). Well, if there's value on the one side, and the ability to deliver under cost and under budget on the other, then that's a value-to-cost asymmetry and subsequently a candidate for a scalable service business...

https://en.wikipedia.org/wiki/Business_process_re-engineerin...

Yes. As someone else points out: the techniques for this exist (and have existed for decades). It was never worth it to fully pursuit it, especially for more messy human-heavy processes.

Now with AI you can get way more detailed (AI can interview humans, and you can do it in a format that doesn't feel like an interview - e.g by 'simply' having AI be a fly on the wall). AI can make sense of messy inputs and then you can present that to people who know the process, who can easily point out flaws. The flows/maps will not be perfect, but you can always get more detailed once you bump into the limitations of imperfection.

And absolutely do business leaders want maps like this. They often have no idea what exactly is happening in their org. Or they have a gut feel that there are massive inefficiencies, but they cannot get the insights into where they are because there are too many moving parts that do not effectively talk to each other (silos, politics, etc).

So, full Panopticon it is then.

I don't think the atomic task based view it the right way to draw conclusions about coming business automation. There is more likely thresholds, after which it makes sense to automate entire sections, or business processes. Creating strange hybrid working situations with Automation foisted on people, will create both the overhead of the IT systems, and the inefficiencies of people trying to deal with edgecase accuracy problems and the like, or lack of skill development paths etc.

Companies are just a collection of people.

The internet is just a series of tubes.

>If you were to have done this exercise for any business in 2022 it would have given that business a significant advantage.

Hmm, what about businesses that have done it in 1922? Because that's how old the methodology is:

https://en.wikipedia.org/wiki/Business_process_mapping

At best this is limited to the algorithmic component of companies that drives information exchange. Beyond it, companies also are a collection of datasets, finances, property, and physical workers affecting property, although yes, they can in principle be controlled by AI. Another part is the evolution of a company's algorithms.

I actually wrote something very similar to this concept a few days ago about what the next steps of AI in software development might look like: https://gpeake.com/blog/abstraction

Wow, things operate according to lists of instructions. What a concept.

Reply on news.ycombinator.com